DEV Community

Cover image for My Messy Journey Building AutoBoss with Bolt.new
Techiral
Techiral

Posted on

My Messy Journey Building AutoBoss with Bolt.new

WLH Challenge: Building with Bolt Submission

This is a submission for the World's Largest Hackathon Writing Challenge: Building with Bolt.

Hey, I’m Lakshya, an 18-year-old high school senior who’s been messing around with code and AI for a while now. I’ve built some AI apps before, but most of the tools out there are either super expensive or made for people with PhDs. When I saw the World’s Largest Hackathon, I thought, “Why not make something for normal people?” That’s where AutoBoss came from—a toolkit to help non-coders use AI for small businesses. I used Bolt.new to pull it off, and let me tell you, it was a wild, stressful ride. Here’s what happened, bugs and all.

What’s AutoBoss?

AutoBoss is for people who don’t code but want AI to help their business. It’s got a few pieces:

  • A place to organize client projects so you don’t lose track.
  • A tool to build AI helpers, like chatbots or something to sort data.
  • A way to train AI with PDFs or websites.
  • A little marketplace to share or sell your AI setups.
  • A guide with steps to find your first client.

It’s not perfect—there’s this one bug where uploading contacts just breaks everything, and I didn’t have time to make it look super slick. But it works, mostly, and I’m kinda proud of that.

Bolt.new and Me Trying to Keep Up

I’ve coded before, so I’m not totally clueless, but Bolt.new was new to me and a total lifesaver. It’s this no-code platform where you just command and vibe code to build entire apps and prototypes without writing a ton of code. I was skeptical at first, like, “Is this actually gonna work?” But it did. For the AI helper part, I just wrote my ideas and creativity to set up basic prototypes. It was stupidly fast.

One thing I got excited about was making a tool to fetch the data from the URL & files, and actually convert that into usable data. I told Bolt.new to implement the proper RAG based on how most of the data was in different documentation. But this didn't work because Netlify and Vercel are just hosting-like services, not actual server providers. And, I had only 2 options left- 1. Use Firecrawl API 2. Buy a server because tools like Puppeteer, etc., need a whole server. I was like, “Holy crap, I don't wanna spend money! I want to make it free for me as the owner and free for all the users.” I didn’t keep any fancy prompt because I was rushing, but it was something like telling it to create its own fetcher that fetches the URL and reads the docs. Bolt.new made that way easier than coding it from scratch. But that wasn't perfect. That had a lot of bugs, and also, the data extraction was inaccurate. So, I needed to open up VS Code on my laptop and did a git clone for my repo. I fixed the extraction and used some libraries to create a mini-clone of that thing. Yeah, that wasn't perfect, but it worked as a perfect jugaad. Then, I pushed the changes to GitHub again and pulled back in the Bolt.new.

My Knowledge Page was Working Properly.

I used Gemini and Genkit for the AI stuff and the main Bolt.new problem was that it was cleaning up the whole .env again and again. Then, I implemented other features also like Agent Personality, Proper Twillio Integration, SendGrid API Integration, etc. Also, I tried to integrate ElevenLabs API keys, but via my creator pack coupon code, the premium pack was not eligible. So, I skipped the plan of using Eleven Labs and stuck to the basic browser or device-based audio. I planned for the Video Calling Agent Builder, but that can't be completed. Because the time was running out. So, I skipped that also and went for the custom domain. I asked Bolt.new to provide me the link for the free custom domain via Entri under the hackathon. And then spent the whole 3-4 hours just guessing the perfect domain name. So, finally, I finalized on the name of theautoboss.org. But when entering the address, I got stuck again. Really, I was irritated now, because the time was running out. And, I was getting too many hurdles. So, for the solution, I went to the Discord server. I asked the query and waited for 3 hours. And I got a reply. They told me to enter a random US address there, but I got stuck there. But after 3-5 tries, I got the final solution, and got to the final page where I entered my debit card, and it failed, and I left my plan to buy a domain (even for free), because I have no credit card and nor do my parents. And congrats, I spent my 3-8 hours on unproductive tasks.

So, when only 2 days were left. I started branding and deploying to Netlify and Vercel. I chose Vercel also because I personally prefer it more. But as per hackathon guidelines, I also used Netlify. Then refined the UI with a dark theme and a minimalistic black UI and asked some friends to use this. But no one was interested in AI. As expected from the Bareilly typical environment. So yeah, I completed the presentation. For time-saving, I used the gamma.app for the presentation creation and fixed the GitHub readme.md for better UI and presentation.

I also used some other stuff:

  • Firebase to store data and handle logins—it was a headache half the time.
  • Supabase for database stuff when Firebase was being a jerk.
  • Next.js and React for the front end, which I know a bit, but still messed up a lot.
  • Hosted it on Vercel and Netlify .
  • Supademo for quick guides to help users not get lost.

Bolt.new was the real hero, though. It let me focus on the app instead of getting stuck in code hell. But, there is a problem, I think Bolt.new is very good as a teammate or a prototyper. But, if I have to create a whole application that can be converted into a real SaaS in just 1 month, then I will need a lot of brain neurons having skills of prompt engineering or a time machine to slow time. Don't mind, but as a student, it was too hard for me to manage my studies and also depend on Bolt.new whole. So for debugging, I need to code some parts on my own. Otherwise, Bolt.new is really like a magical Doraemon gadget that can turn ideas into reality.

Where It All Fell Apart

Building this by myself was brutal. I started in June, and the July deadline was looming, so I was up until 3 a.m. most nights, drinking energy drinks and stressing. No team meant I had to do everything—design, code, test, fix bugs. I’m used to coding, but doing all these new tools while racing the clock was very tough.

The tech was a nightmare sometimes. Firebase and Supabase kept throwing errors because they weren’t syncing right, and I spent way too long on Stack Overflow trying to fix it. That contacts upload bug? It’s still there, mocking me every time it crashes. The worst was two days before the deadline when I realized I had already used all the credentials in the chat of the Bolt.new. So, I made that project private and created a new Bolt.new project by importing that GitHub repo, which excludes every primary credential. And, I submitted that one in the hackathon (not the original one).

The Good Stuff

I’m still kinda shocked it works. My friend’s mom, who runs a bakery, used the chatbot I set up, and she was like, “Lakshya, this thing answers customer questions faster than I do!” That felt better than anything. Building something this big on my own, especially with tools I’d never used before, makes me think I might actually be able to do this coding thing for real.

What I Figured Out

I learned that non-coders need things to be super simple. I added a guide with clear steps and some dumb badges for fun, and my testers were like, “Okay, this makes sense now.” I also got way better at Next.js and Supabase, which scared me at first. Bolt.new showed me I don’t always have to code every little thing—sometimes you just need a tool that works.

What’s Next?

I want to fix that contacts bug—it’s driving me nuts. Maybe add more AI templates, like integrating with Slack, Trello, Make.com, N8N, and Zapier for broader business use, or expanding AI templates for CRM, social media automation, and lead generation. If people like it, I’d love to turn AutoBoss into something real, but I’m not getting my hopes up too high. I’m just a kid with a laptop and a million ideas, you know?

Why I Care

AutoBoss isn’t some shiny startup app, but I love that it makes AI less scary for regular people like my friend’s mom. The hackathon pushed me to my limits, and I’m glad I didn’t quit, even when I wanted to. I’m excited to keep tinkering—maybe fix those bugs, maybe dream a little bigger.

Built with too many energy drinks and a lot of yelling at my laptop.

That’s my story of building AutoBoss with Bolt.new. It’s a mess, it’s got bugs, but it’s mine. Hope it shows what a high school coder can pull off when they’re stressed but stubborn. Thanks for reading!

Top comments (0)